Can I run Laguna-S-2.1 on a Apple M1 Pro?
Not at these settings. No indexed quantization of Laguna-S-2.1 fits Apple M1 Pro at any context we compute, with q8_0 KV. The smallest shipped quantization is 23.15 GiB in weights alone, against 11.16 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 219.05 GiB | 219.8 | 219.9 | 220.1 | 220.5 | 221.3 | 222.9 |
| F16 | 219.05 GiB | 219.8 | 219.9 | 220.1 | 220.5 | 221.3 | 222.9 |
| Q8_0 | 119.91 GiB | 120.7 | 120.8 | 121.0 | 121.4 | 122.2 | 123.7 |
| Q6_K_L | 94.83 GiB | 95.6 | 95.7 | 95.9 | 96.3 | 97.1 | 98.7 |
| UD-Q6_K | 91.19 GiB | 91.9 | 92.0 | 92.2 | 92.6 | 93.4 | 95.0 |
| Q6_K | 89.93 GiB | 90.7 | 90.8 | 91.0 | 91.4 | 92.2 | 93.8 |
| Q4_K_M | 89.44 GiB | 90.2 | 90.3 | 90.5 | 90.9 | 91.7 | 93.3 |
| UD-Q5_K_M | 81.83 GiB | 82.6 | 82.7 | 82.9 | 83.3 | 84.1 | 85.7 |
| Q5_K_L | 78.39 GiB | 79.1 | 79.2 | 79.4 | 79.8 | 80.6 | 82.2 |
| Q5_K_M | 78.22 GiB | 79.0 | 79.1 | 79.3 | 79.7 | 80.5 | 82.1 |
| UD-Q5_K_S | 76.98 GiB | 77.7 | 77.8 | 78.0 | 78.4 | 79.2 | 80.8 |
| Q5_K_S | 75.72 GiB | 76.5 | 76.6 | 76.8 | 77.2 | 78.0 | 79.6 |
| Q4_1 | 68.96 GiB | 69.7 | 69.8 | 70.0 | 70.4 | 71.2 | 72.8 |
| UD-Q4_K_M | 68.10 GiB | 68.8 | 68.9 | 69.1 | 69.5 | 70.3 | 71.9 |
| Q4_K_S | 64.36 GiB | 65.1 | 65.2 | 65.4 | 65.8 | 66.6 | 68.2 |
| UD-Q4_K_S | 63.88 GiB | 64.6 | 64.7 | 64.9 | 65.3 | 66.1 | 67.7 |
| Q4_0 | 62.44 GiB | 63.2 | 63.3 | 63.5 | 63.9 | 64.7 | 66.3 |
| IQ4_NL | 62.32 GiB | 63.1 | 63.2 | 63.4 | 63.8 | 64.6 | 66.2 |
| IQ4_XS | 58.98 GiB | 59.7 | 59.8 | 60.0 | 60.4 | 61.2 | 62.8 |
| UD-IQ4_NL | 54.71 GiB | 55.5 | 55.6 | 55.8 | 56.2 | 57.0 | 58.6 |
| UD-IQ4_XS | 53.61 GiB | 54.4 | 54.5 | 54.7 | 55.1 | 55.9 | 57.4 |
| IQ3_M | 52.60 GiB | 53.3 | 53.4 | 53.6 | 54.0 | 54.8 | 56.4 |
| Q3_K_L | 52.33 GiB | 53.1 | 53.2 | 53.4 | 53.8 | 54.6 | 56.2 |
| Q3_K_M | 50.33 GiB | 51.1 | 51.2 | 51.4 | 51.8 | 52.6 | 54.2 |
| UD-Q3_K_M | 50.31 GiB | 51.1 | 51.2 | 51.4 | 51.8 | 52.6 | 54.1 |
| IQ3_XS | 50.29 GiB | 51.0 | 51.1 | 51.3 | 51.7 | 52.5 | 54.1 |
| Q3_K_S | 48.03 GiB | 48.8 | 48.9 | 49.1 | 49.5 | 50.3 | 51.9 |
| IQ3_XXS | 46.07 GiB | 46.8 | 46.9 | 47.1 | 47.5 | 48.3 | 49.9 |
| UD-IQ3_S | 45.10 GiB | 45.9 | 46.0 | 46.1 | 46.5 | 47.3 | 48.9 |
| UD-IQ3_XXS | 41.24 GiB | 42.0 | 42.1 | 42.3 | 42.7 | 43.5 | 45.1 |
| Q2_K_L | 39.20 GiB | 39.9 | 40.0 | 40.2 | 40.6 | 41.4 | 43.0 |
| Q2_K | 38.92 GiB | 39.7 | 39.8 | 40.0 | 40.4 | 41.2 | 42.8 |
| IQ2_M | 37.19 GiB | 37.9 | 38.0 | 38.2 | 38.6 | 39.4 | 41.0 |
| UD-IQ2_M | 34.71 GiB | 35.5 | 35.6 | 35.8 | 36.2 | 37.0 | 38.5 |
| UD-IQ2_XXS | 34.64 GiB | 35.4 | 35.5 | 35.7 | 36.1 | 36.9 | 38.5 |
| IQ2_S | 33.77 GiB | 34.5 | 34.6 | 34.8 | 35.2 | 36.0 | 37.6 |
| UD-IQ1_M | 33.19 GiB | 33.9 | 34.0 | 34.2 | 34.6 | 35.4 | 37.0 |
| IQ2_XS | 33.12 GiB | 33.9 | 34.0 | 34.2 | 34.6 | 35.4 | 37.0 |
| UD-IQ1_S | 31.45 GiB | 32.2 | 32.3 | 32.5 | 32.9 | 33.7 | 35.3 |
| IQ2_XXS | 29.82 GiB | 30.6 | 30.7 | 30.9 | 31.3 | 32.1 | 33.7 |
| IQ1_M | 25.75 GiB | 26.5 | 26.6 | 26.8 | 27.2 | 28.0 | 29.6 |
| IQ1_S | 23.15 GiB | 23.9 | 24.0 | 24.2 | 24.6 | 25.4 | 27.0 |
Figures are GiB of total memory: weights plus KV cache plus compute buffer and backend overhead. Weights and KV are near-exact; the overhead term is modeled. Hover any cell for the breakdown.
Why other calculators disagree
A parameters × bits ÷ 8 estimate ignores two things that dominate at long context. First, the weights themselves are not the nominal rate — quantizations are mixtures, so the real file is consistently larger than the label implies. Second, most of this model's layers cache only a 512-token window rather than the full context.